Special Issue of the Journal “Pattern Recognition”
Analysis and Recognition of Indirect Immunofluorescence Images
Call for Papers
In the recent years we have assisted to a progressive growing number of Pattern Recognition applications, mainly devoted to the exploitation of cutting edge scientific methodologies for the solution of problems of relevant interest to civil society. In the field of medical image analysis this trend has been even more evident than in other ones, as the availability of assisted diagnosis tools would allow the medical community to increase their productivity jointly with an improvement of the quality and precision of the diagnostic act.
Among all, rather novel interests are concentrating on the indirect immunofluorescence images (IIF), i.e. images obtained by “staining” a biological tissue with antibodies that carry a fluorescent chemical compound and that bind to other specific antibodies, so as to generate fluorescence images representing the distribution of the target antibodies within the tissue. This kind of assay is considered the gold standard to evaluate the presence of several autoimmune diseases. Due to its effectiveness, diagnostic tests for systemic autoimmune diseases are now becoming more and more interesting to industrial communities; there is a consequently strong demand for a complete automation of the process that would result in increased repeatability and reliability, easier and faster result reporting, higher productivity and lower costs.
The automation of this process is a topic that is meeting a growing interest among PR scientists, that in the last few years have provided innovative contributions to the different aspects of the analysis of these kind of images; a clear evidence of this interest has been recently demonstrated in occasion of the first edition of the “HEp‑2 cells classification” contest hosted by the ICPR 2012. The initiative got a large participation of the scientific community being the ICPR 2012 contest with the highest number of submissions. In light of this interest and considering that, up to now, no journal has dedicated a special issue to this topic, it appears evident the need of presenting the state of the art of these emerging applied Pattern Recognition systems. The occasion of connecting this special issue with the contest is particularly important: it would be not merely an issue devoted to collect papers on a single interesting topic, but also the unmissable opportunity to assess the performance of different methods, by comparing them on the common database used for the ICPR 2012 contest.
The special issue aims at gathering contributions analyzing the state of the art in this field and describing the ongoing research activities by inviting manuscripts on the most recent developments in this applicative area, and welcoming those validated on the dataset used for the ICPR 2012 contest (see mivia.unisa.it/contest-hep-2/ ). Topics of interest include, but are not limited to:
- Image acquisition and pre-processing
- Cells segmentation
- Mitotic cell detection
- Image fluorescence intensity classification
- Image staining pattern recognition
- CAD systems based on IFI images
Submissions to the special issue must include new, unpublished, original research. Papers must be original and have not been published or submitted elsewhere. All papers must be written in English. The submissions will be reviewed by at least three reviewers. Before submission authors should carefully read over the Journal Instructions for Authors.
Prospective authors should submit an electronic copy of their complete manuscript through the Pattern Recognition submission system at http://ees.elsevier.com/pr/, choosing the “Special Issue: IIF Image Analysis” in the Article Field Type.
In order to promote the direct comparison of the obtained performance with the results of the other methods, including the ones participating to the 2012 HEp-2 Cells Classification Contest, authors are required to provide in their papers all the performance indicators detailed below. Authors are free to supply also other performance measures, if they desire, but they cannot omit any of the required measures; non-compliance with this requirement will be considered a reason for exclusion from the special issue.A first experimentation must be performed dividing the images into a training and a test set in exactly the same way as it was done in the 2012 HEp-2 Cells Classification Contest. The authors must train their algorithms using the images provided as training set in the HEp-2 Contest, and must test the algorithms using the cells from the images used as test set for the contest. The classification must evaluate each cell individually; it must not use information on the other cells belonging to the same image.The results of this experimentation must be presented as follows:
- the authors must present the cell-level classification confusion matrix, a table where the entry corresponding to row i and column j represents the percentage of cells of class i assigned to class j with respect to the total number of cells of class i in the test set;
- the authors must present the average cell-level accuracy, i.e. the overall percentage of cells that are correctly classified.
Furthermore, the authors have to compute the above information also at the image-level, using as the guessed class for each image the most frequently assigned class to the cells within that image:
- the image-level classification confusion matrix, representing the percentage of images of class i assigned to class j with respect to the total number of images in the test set;
- the image-level accuracy, i.e. the overall percentage of images that are correctly classified.
Please take note that BOTH the cell-level and the image-level measures are required.
A second experimentation must be performed using the leave-one-out technique over all the 28 images: for each image in the dataset, a classifier instance must be trained using the other 27 images; this classifier must be used to classify the cells of the image left out. Also here the evaluation must not use information on the other cells belonging to the same image. Please note that BOTH the first experimentation and this second one must be performed; they are not in alternative.
The results of this second experimentation must be presented as follows:
- a table with a row for each image, showing the true class of the image, and for each class the number and the percentage of cells in that image that have been assigned to the class;
- the cell-level confusion matrix and average accuracy, obtained by adding the classification counts of the 28 leave-one-out runs: the number of cells of class i assigned to class j must be computed by summing the corresponding numbers in each run, and then computing the percentages (and not by averaging the percentages relative to each run);
- the image-level confusion matrix and average accuracy, obtained by using as guessed class the most frequently assigned class of the cells within the image, and adding the resulting classification counts of the 28 leave-one-out runs.
|Manuscript submission deadline:|
|Revised manuscript submission:|
|Notification of final decision:|
|Planned publication date of special issue:||Winter 2013|
Pasquale Foggia1, Gennaro Percannella1, Paolo Soda2, Mario Vento1
Corresponding Guest Editor: Prof. Mario Vento, E-mail: email@example.com
1: Department of Electronic and Computer Engineering, University of Salerno, Italy.
2: Medical Informatics and Computer Science Laboratory, Integrated Research Centre, University Campus Bio-Medico of Rome, Italy.